Future of CAD / generative design article

– Sec 1

What will industrial design look like in the future ? Insights from computer science and the maker movement give us a few hints.

The last sea change in CAD for product design was the appearance of Solidworks on the scene, back in 1995. The incumbent AutoCAD had not evolved too much over the previous ~10 years, since offering a 2D drafting experience which was inspired by paper drafting, a world of lines, curves and planes, which was adequate in the early PC era of the 1980’s and early 1990’s.

Solidworks allowed the field to catch up with the hardware that became available in the late 1990’s, by offering a full 3D experience, a parametric model of object designs, assemblies, relationships and constraints between those objects, etc.

With Solidworks, the CAD world transitioned from drafting to modeling.

I am proposing/asserting that we may well be at another inflection point. [departure/paradigm change/…] However, as was the case in the past, the incumbents of today, ie Solidworks, AutoDesk, Creo, … are not going to be driving those changes. But this time, no outside disruptor seems to be about to appear on the scene. AutoDesk* and Solidworks are making good profit from releasing new updates every year, with only gradual innovation. A startup called Onshape is offering an online alternative to Solidworks, but apart from this, seems to be proposing pretty much the same parametric modelling experience.

What is the change we would like to see ?

– Sec 2

The first of those disruptions is one that manufacturing companies could decide to bring about. Right now, there is little integration between CAD and manufacturing. What if this were to change? What if CAD software had a ‘Make’ button ?

Of course, we are not suggesting that any imaginable product design should be available for ordering and shipping by a ‘One-Click order’. But such a functionality could be made available within certain product niches. In fact, we are aware of a few examples already :

Still, we think there is further progress to be made in this direction. For a start, CAD software should be at least able to generate not just a BOM (bill of materials), but say a Farnell or McMaster order* for the stock parts in the design. A trickier step could be to propose automatic representative quotes from local toolmakers for any machined metal parts in an assembly. Those things may be implementable as Solidworks plugins, if they aren’t already in part.

Finally, one would imagine manufacturer provided plugins or standalone software for specialized part designs such as pressings and [find other example]

– Sec 3

The current hegemony of parametric modelling might lead one to believe that few, if any alternatives might ever be able to emerge. That need not be the case. A particular class of design paradigm called generative design might become ever more viable in the future, at least in some areas of product design. This belief is brought about by the continued increase in computing power, as well as recent progress in fundamental techniques in constraint solvers, knowledge modelling, and other fields.

We can look at parametric modelling as a “bottom-up” process, where the model is built from low-level, primitive relations [constraints] of the sort “surfaces A and B are parallel, and 10mm apart”, “ a 10-inch radius hole in the middle of the front face C”, “a 100mm beam located at X, Y, Z”, etc.

_Generative design” is the name we choose to give to a top-down, high-level approach. Generally, as in parametric modelling, a model is built with statements representing a constraint, or describing an object. However, the difference between generative design and parametric modelling is that these are much higher-level constraints.

An example of such constraint statements : “A gearbox to fit a 100mmx40mmx40mm space, has 5 gears and such and such power characteristics”. Or even more general : “A white chair, Eames style”

In generative design, we need a process to synthetize (generate) a low-level model from these high-level statements.


There is not currently a lot of research in these areas, apart from a somewhat different problem called “topology optimization” which is getting attention from some research groups, such as AutoDesk.

– Sec 4 : organising information / knowledge model

Another improvement to the world of product design would be to provide better access to domain knowledge. Designing a product requires a lot of domain-specific information, which specialists take a long time to learn and integrate : in a given product category, what are the main design options ? The best design choices for various projected uses ? the best materials to use ? What design approach to take ? What build processes are known ? What styles ? etc. In this day and age, we still mostly rely on the skills and knowledge of the individual designer to answer those questions.

Easier and more systematic access to domain knowledge would help further democratise product design.

This is not a new problem, anvd various approaches have been developed over the years. Solutions available today can be understood to exists at two ends of the design space for knowledge engineering.

On the one end, designers and engineers look for information on general-purpose sites like Google or Wikipedia. They are accessible to all, but will often only offer low-quality information. It is possible to find inspiration or even useful information, but not in a systematic way. There are a few more specialized databases for parts such as ThomasNet, but you need to know what you are looking for.

At the other end of the scale is what is know under the general term of “Knowledge-Based Engineering” (KBE). KBE is a vague term that can be understood in various ways. Here, we take it to mean an effort to integrate domain specific design rules into CAD software.

Typically, KBE implementations augment the CAD platform with a scripting language for the user to program the generation of a model from a set of parameters. For instance, a company used a KBE system to generate a range of windscreen wipers to fit all the various vehicles it was working with.

The initial hype around KBE slowed down in the late 1980s, as part of the “AI Winter” around that time. This mens that there has been comparatively little research into KBE since the Web appeared. Between general search engines, and KBE-style systems, we feel that there is still a huge space for innovation to better organise design knowledge. For instance, designers would benefit from being able to consult a wide-ranging public “tech tree” (not literally a tree ; an “ontology” in technical terms) to provide access to design knowledge that is searchable, but also as systematic and complete as possible.

[Or : We think there is the potential to take the best of search engines, wikis, KBE systems, and research into knowledge models, to provide much wider access to a much larger corpus design knowledge.]